In context learning is learning. Then you bunch up things, and the next time your computer goes to sleep it finetunes on it.
@karpathy
-
AI Model Capability to Simulate Doom Game
By
–
It can probably very close to simulate Doom if you ask nicely.
-
Pure Software 2.0: Neural Network-Only Computer Architecture
By
–
100% Fully Software 2.0 computer. Just a single neural net and no classical software at all. Device inputs (audio video, touch etc) directly feed into a neural net, the outputs of it directly display as audio/video on speaker/screen, that’s it.
-
On-Device Health Check for Phone Security and Privacy
By
–
You’re preaching to the choir I think there could be fully secure and private ways to do this in an automated, on device way. It’s like a part of the phone has a “health check” estimating whether it is being used statistically in normal way or if it was “abducted” like this.
-
Physical Devices With Sensors Hold Significant Contextual Advantages
By
–
I just feel like a physical device with so many sensors and so much context over time has a very very high advantage should it wish to use it. Esp if it’s done in a secure element or so.
-
AI-Generated Fake People for Hire Will Become Indistinguishable
By
–
The “attack” will shortly get significantly more sophisticated and indistinguishable from human wrt what we’ve seen so far. These will look like real people but fully fake and for hire.
-
On-Device ML Liveness Detection and Cryptographic Certification for Real Users
By
–
Could iOS/Android OS do some kind of on-device ML for liveness detection and securely, privately cryptographically sign/certify actions as coming from a live, real person?
-
llm.c Memory Efficiency Advantages Over PyTorch
By
–
If you match the parameters you're actually under-estimating the improvement, because llm.c takes up a lot less space so you can crank up the batch size. I haven't fully dug into why PyTorch takes up that much space, slightly worried I'm doing something wrong but not sure what
-
Karpathy highlights extraordinary moment in AI at World’s Fair
By
–
The @aiDotEngineer World's Fair in SF this week https://
ai.engineer/worldsfair Reminded of slide #1 from my most recent talk: "Just in case you were wondering…
No, this is not a normal moment in AI" -
MLX Framework as Optimized Alternative to PyTorch MPS
By
–
I'd have a look at MLX, cc @awnihannun , from some recent chatter I understand it is a lot more optimized that PyTorch mps atm. This would need a re-write of the build-nanogpt code into mlx, but possibly it's not that involved. I'd be happy to accept a PR for mlx clone.